Commit | Line | Data |
---|---|---|
0793a61d TG |
1 | /* |
2 | * Performance counters: | |
3 | * | |
4 | * Copyright(C) 2008, Thomas Gleixner <tglx@linutronix.de> | |
5 | * Copyright(C) 2008, Red Hat, Inc., Ingo Molnar | |
6 | * | |
7 | * Data type definitions, declarations, prototypes. | |
8 | * | |
9 | * Started by: Thomas Gleixner and Ingo Molnar | |
10 | * | |
11 | * For licencing details see kernel-base/COPYING | |
12 | */ | |
13 | #ifndef _LINUX_PERF_COUNTER_H | |
14 | #define _LINUX_PERF_COUNTER_H | |
15 | ||
f3dfd265 PM |
16 | #include <linux/types.h> |
17 | #include <linux/ioctl.h> | |
9aaa131a | 18 | #include <asm/byteorder.h> |
0793a61d TG |
19 | |
20 | /* | |
9f66a381 IM |
21 | * User-space ABI bits: |
22 | */ | |
23 | ||
24 | /* | |
b8e83514 | 25 | * hw_event.type |
0793a61d | 26 | */ |
b8e83514 PZ |
27 | enum perf_event_types { |
28 | PERF_TYPE_HARDWARE = 0, | |
29 | PERF_TYPE_SOFTWARE = 1, | |
30 | PERF_TYPE_TRACEPOINT = 2, | |
31 | ||
0793a61d | 32 | /* |
b8e83514 | 33 | * available TYPE space, raw is the max value. |
0793a61d | 34 | */ |
9f66a381 | 35 | |
b8e83514 PZ |
36 | PERF_TYPE_RAW = 128, |
37 | }; | |
6c594c21 | 38 | |
b8e83514 PZ |
39 | /* |
40 | * Generalized performance counter event types, used by the hw_event.event_id | |
41 | * parameter of the sys_perf_counter_open() syscall: | |
42 | */ | |
43 | enum hw_event_ids { | |
9f66a381 | 44 | /* |
b8e83514 | 45 | * Common hardware events, generalized by the kernel: |
9f66a381 | 46 | */ |
b8e83514 PZ |
47 | PERF_COUNT_CPU_CYCLES = 0, |
48 | PERF_COUNT_INSTRUCTIONS = 1, | |
49 | PERF_COUNT_CACHE_REFERENCES = 2, | |
50 | PERF_COUNT_CACHE_MISSES = 3, | |
51 | PERF_COUNT_BRANCH_INSTRUCTIONS = 4, | |
52 | PERF_COUNT_BRANCH_MISSES = 5, | |
53 | PERF_COUNT_BUS_CYCLES = 6, | |
54 | ||
55 | PERF_HW_EVENTS_MAX = 7, | |
56 | }; | |
e077df4f | 57 | |
b8e83514 PZ |
58 | /* |
59 | * Special "software" counters provided by the kernel, even if the hardware | |
60 | * does not support performance counters. These counters measure various | |
61 | * physical and sw events of the kernel (and allow the profiling of them as | |
62 | * well): | |
63 | */ | |
64 | enum sw_event_ids { | |
65 | PERF_COUNT_CPU_CLOCK = 0, | |
66 | PERF_COUNT_TASK_CLOCK = 1, | |
67 | PERF_COUNT_PAGE_FAULTS = 2, | |
68 | PERF_COUNT_CONTEXT_SWITCHES = 3, | |
69 | PERF_COUNT_CPU_MIGRATIONS = 4, | |
70 | PERF_COUNT_PAGE_FAULTS_MIN = 5, | |
71 | PERF_COUNT_PAGE_FAULTS_MAJ = 6, | |
72 | ||
73 | PERF_SW_EVENTS_MAX = 7, | |
0793a61d TG |
74 | }; |
75 | ||
76 | /* | |
77 | * IRQ-notification data record type: | |
78 | */ | |
9f66a381 | 79 | enum perf_counter_record_type { |
b8e83514 PZ |
80 | PERF_RECORD_SIMPLE = 0, |
81 | PERF_RECORD_IRQ = 1, | |
82 | PERF_RECORD_GROUP = 2, | |
0793a61d TG |
83 | }; |
84 | ||
f4a2deb4 PZ |
85 | #define __PERF_COUNTER_MASK(name) \ |
86 | (((1ULL << PERF_COUNTER_##name##_BITS) - 1) << \ | |
87 | PERF_COUNTER_##name##_SHIFT) | |
88 | ||
89 | #define PERF_COUNTER_RAW_BITS 1 | |
90 | #define PERF_COUNTER_RAW_SHIFT 63 | |
91 | #define PERF_COUNTER_RAW_MASK __PERF_COUNTER_MASK(RAW) | |
92 | ||
93 | #define PERF_COUNTER_CONFIG_BITS 63 | |
94 | #define PERF_COUNTER_CONFIG_SHIFT 0 | |
95 | #define PERF_COUNTER_CONFIG_MASK __PERF_COUNTER_MASK(CONFIG) | |
96 | ||
97 | #define PERF_COUNTER_TYPE_BITS 7 | |
98 | #define PERF_COUNTER_TYPE_SHIFT 56 | |
99 | #define PERF_COUNTER_TYPE_MASK __PERF_COUNTER_MASK(TYPE) | |
100 | ||
101 | #define PERF_COUNTER_EVENT_BITS 56 | |
102 | #define PERF_COUNTER_EVENT_SHIFT 0 | |
103 | #define PERF_COUNTER_EVENT_MASK __PERF_COUNTER_MASK(EVENT) | |
104 | ||
9f66a381 IM |
105 | /* |
106 | * Hardware event to monitor via a performance monitoring counter: | |
107 | */ | |
108 | struct perf_counter_hw_event { | |
f4a2deb4 PZ |
109 | /* |
110 | * The MSB of the config word signifies if the rest contains cpu | |
111 | * specific (raw) counter configuration data, if unset, the next | |
112 | * 7 bits are an event type and the rest of the bits are the event | |
113 | * identifier. | |
114 | */ | |
115 | __u64 config; | |
9f66a381 | 116 | |
f3dfd265 | 117 | __u64 irq_period; |
2743a5b0 PM |
118 | __u64 record_type; |
119 | __u64 read_format; | |
9f66a381 | 120 | |
2743a5b0 | 121 | __u64 disabled : 1, /* off by default */ |
0475f9ea | 122 | nmi : 1, /* NMI sampling */ |
0475f9ea PM |
123 | inherit : 1, /* children inherit it */ |
124 | pinned : 1, /* must always be on PMU */ | |
125 | exclusive : 1, /* only group on PMU */ | |
126 | exclude_user : 1, /* don't count user */ | |
127 | exclude_kernel : 1, /* ditto kernel */ | |
128 | exclude_hv : 1, /* ditto hypervisor */ | |
2743a5b0 | 129 | exclude_idle : 1, /* don't count when idle */ |
0475f9ea | 130 | |
b8e83514 | 131 | __reserved_1 : 55; |
2743a5b0 PM |
132 | |
133 | __u32 extra_config_len; | |
134 | __u32 __reserved_4; | |
9f66a381 | 135 | |
f3dfd265 | 136 | __u64 __reserved_2; |
2743a5b0 | 137 | __u64 __reserved_3; |
eab656ae TG |
138 | }; |
139 | ||
d859e29f PM |
140 | /* |
141 | * Ioctls that can be done on a perf counter fd: | |
142 | */ | |
143 | #define PERF_COUNTER_IOC_ENABLE _IO('$', 0) | |
144 | #define PERF_COUNTER_IOC_DISABLE _IO('$', 1) | |
145 | ||
37d81828 PM |
146 | /* |
147 | * Structure of the page that can be mapped via mmap | |
148 | */ | |
149 | struct perf_counter_mmap_page { | |
150 | __u32 version; /* version number of this structure */ | |
151 | __u32 compat_version; /* lowest version this is compat with */ | |
152 | __u32 lock; /* seqlock for synchronization */ | |
153 | __u32 index; /* hardware counter identifier */ | |
154 | __s64 offset; /* add to hardware counter value */ | |
155 | }; | |
156 | ||
f3dfd265 | 157 | #ifdef __KERNEL__ |
9f66a381 | 158 | /* |
f3dfd265 | 159 | * Kernel-internal data types and definitions: |
9f66a381 IM |
160 | */ |
161 | ||
f3dfd265 PM |
162 | #ifdef CONFIG_PERF_COUNTERS |
163 | # include <asm/perf_counter.h> | |
164 | #endif | |
165 | ||
166 | #include <linux/list.h> | |
167 | #include <linux/mutex.h> | |
168 | #include <linux/rculist.h> | |
169 | #include <linux/rcupdate.h> | |
170 | #include <linux/spinlock.h> | |
d6d020e9 | 171 | #include <linux/hrtimer.h> |
f3dfd265 PM |
172 | #include <asm/atomic.h> |
173 | ||
174 | struct task_struct; | |
175 | ||
f4a2deb4 PZ |
176 | static inline u64 perf_event_raw(struct perf_counter_hw_event *hw_event) |
177 | { | |
178 | return hw_event->config & PERF_COUNTER_RAW_MASK; | |
179 | } | |
180 | ||
181 | static inline u64 perf_event_config(struct perf_counter_hw_event *hw_event) | |
182 | { | |
183 | return hw_event->config & PERF_COUNTER_CONFIG_MASK; | |
184 | } | |
185 | ||
186 | static inline u64 perf_event_type(struct perf_counter_hw_event *hw_event) | |
187 | { | |
188 | return (hw_event->config & PERF_COUNTER_TYPE_MASK) >> | |
189 | PERF_COUNTER_TYPE_SHIFT; | |
190 | } | |
191 | ||
192 | static inline u64 perf_event_id(struct perf_counter_hw_event *hw_event) | |
193 | { | |
194 | return hw_event->config & PERF_COUNTER_EVENT_MASK; | |
195 | } | |
196 | ||
0793a61d | 197 | /** |
9f66a381 | 198 | * struct hw_perf_counter - performance counter hardware details: |
0793a61d TG |
199 | */ |
200 | struct hw_perf_counter { | |
ee06094f | 201 | #ifdef CONFIG_PERF_COUNTERS |
d6d020e9 PZ |
202 | union { |
203 | struct { /* hardware */ | |
204 | u64 config; | |
205 | unsigned long config_base; | |
206 | unsigned long counter_base; | |
207 | int nmi; | |
208 | unsigned int idx; | |
209 | }; | |
210 | union { /* software */ | |
211 | atomic64_t count; | |
212 | struct hrtimer hrtimer; | |
213 | }; | |
214 | }; | |
ee06094f | 215 | atomic64_t prev_count; |
9f66a381 | 216 | u64 irq_period; |
ee06094f IM |
217 | atomic64_t period_left; |
218 | #endif | |
0793a61d TG |
219 | }; |
220 | ||
221 | /* | |
222 | * Hardcoded buffer length limit for now, for IRQ-fed events: | |
223 | */ | |
9f66a381 | 224 | #define PERF_DATA_BUFLEN 2048 |
0793a61d TG |
225 | |
226 | /** | |
227 | * struct perf_data - performance counter IRQ data sampling ... | |
228 | */ | |
229 | struct perf_data { | |
9f66a381 IM |
230 | int len; |
231 | int rd_idx; | |
232 | int overrun; | |
233 | u8 data[PERF_DATA_BUFLEN]; | |
0793a61d TG |
234 | }; |
235 | ||
621a01ea IM |
236 | struct perf_counter; |
237 | ||
238 | /** | |
239 | * struct hw_perf_counter_ops - performance counter hw ops | |
240 | */ | |
241 | struct hw_perf_counter_ops { | |
95cdd2e7 | 242 | int (*enable) (struct perf_counter *counter); |
7671581f IM |
243 | void (*disable) (struct perf_counter *counter); |
244 | void (*read) (struct perf_counter *counter); | |
621a01ea IM |
245 | }; |
246 | ||
6a930700 IM |
247 | /** |
248 | * enum perf_counter_active_state - the states of a counter | |
249 | */ | |
250 | enum perf_counter_active_state { | |
3b6f9e5c | 251 | PERF_COUNTER_STATE_ERROR = -2, |
6a930700 IM |
252 | PERF_COUNTER_STATE_OFF = -1, |
253 | PERF_COUNTER_STATE_INACTIVE = 0, | |
254 | PERF_COUNTER_STATE_ACTIVE = 1, | |
255 | }; | |
256 | ||
9b51f66d IM |
257 | struct file; |
258 | ||
0793a61d TG |
259 | /** |
260 | * struct perf_counter - performance counter kernel representation: | |
261 | */ | |
262 | struct perf_counter { | |
ee06094f | 263 | #ifdef CONFIG_PERF_COUNTERS |
04289bb9 | 264 | struct list_head list_entry; |
592903cd | 265 | struct list_head event_entry; |
04289bb9 IM |
266 | struct list_head sibling_list; |
267 | struct perf_counter *group_leader; | |
5c92d124 | 268 | const struct hw_perf_counter_ops *hw_ops; |
04289bb9 | 269 | |
6a930700 | 270 | enum perf_counter_active_state state; |
c07c99b6 | 271 | enum perf_counter_active_state prev_state; |
0793a61d | 272 | atomic64_t count; |
ee06094f | 273 | |
9f66a381 | 274 | struct perf_counter_hw_event hw_event; |
0793a61d TG |
275 | struct hw_perf_counter hw; |
276 | ||
277 | struct perf_counter_context *ctx; | |
278 | struct task_struct *task; | |
9b51f66d | 279 | struct file *filp; |
0793a61d | 280 | |
9b51f66d | 281 | struct perf_counter *parent; |
d859e29f PM |
282 | struct list_head child_list; |
283 | ||
0793a61d | 284 | /* |
d859e29f | 285 | * Protect attach/detach and child_list: |
0793a61d TG |
286 | */ |
287 | struct mutex mutex; | |
288 | ||
289 | int oncpu; | |
290 | int cpu; | |
291 | ||
37d81828 PM |
292 | /* pointer to page shared with userspace via mmap */ |
293 | unsigned long user_page; | |
294 | ||
0793a61d TG |
295 | /* read() / irq related data */ |
296 | wait_queue_head_t waitq; | |
297 | /* optional: for NMIs */ | |
298 | int wakeup_pending; | |
299 | struct perf_data *irqdata; | |
300 | struct perf_data *usrdata; | |
301 | struct perf_data data[2]; | |
592903cd | 302 | |
e077df4f | 303 | void (*destroy)(struct perf_counter *); |
592903cd | 304 | struct rcu_head rcu_head; |
ee06094f | 305 | #endif |
0793a61d TG |
306 | }; |
307 | ||
308 | /** | |
309 | * struct perf_counter_context - counter context structure | |
310 | * | |
311 | * Used as a container for task counters and CPU counters as well: | |
312 | */ | |
313 | struct perf_counter_context { | |
314 | #ifdef CONFIG_PERF_COUNTERS | |
315 | /* | |
d859e29f PM |
316 | * Protect the states of the counters in the list, |
317 | * nr_active, and the list: | |
0793a61d TG |
318 | */ |
319 | spinlock_t lock; | |
d859e29f PM |
320 | /* |
321 | * Protect the list of counters. Locking either mutex or lock | |
322 | * is sufficient to ensure the list doesn't change; to change | |
323 | * the list you need to lock both the mutex and the spinlock. | |
324 | */ | |
325 | struct mutex mutex; | |
04289bb9 IM |
326 | |
327 | struct list_head counter_list; | |
592903cd | 328 | struct list_head event_list; |
0793a61d TG |
329 | int nr_counters; |
330 | int nr_active; | |
d859e29f | 331 | int is_active; |
0793a61d TG |
332 | struct task_struct *task; |
333 | #endif | |
334 | }; | |
335 | ||
336 | /** | |
337 | * struct perf_counter_cpu_context - per cpu counter context structure | |
338 | */ | |
339 | struct perf_cpu_context { | |
340 | struct perf_counter_context ctx; | |
341 | struct perf_counter_context *task_ctx; | |
342 | int active_oncpu; | |
343 | int max_pertask; | |
3b6f9e5c | 344 | int exclusive; |
96f6d444 PZ |
345 | |
346 | /* | |
347 | * Recursion avoidance: | |
348 | * | |
349 | * task, softirq, irq, nmi context | |
350 | */ | |
351 | int recursion[4]; | |
0793a61d TG |
352 | }; |
353 | ||
354 | /* | |
355 | * Set by architecture code: | |
356 | */ | |
357 | extern int perf_max_counters; | |
358 | ||
359 | #ifdef CONFIG_PERF_COUNTERS | |
5c92d124 | 360 | extern const struct hw_perf_counter_ops * |
621a01ea IM |
361 | hw_perf_counter_init(struct perf_counter *counter); |
362 | ||
0793a61d TG |
363 | extern void perf_counter_task_sched_in(struct task_struct *task, int cpu); |
364 | extern void perf_counter_task_sched_out(struct task_struct *task, int cpu); | |
365 | extern void perf_counter_task_tick(struct task_struct *task, int cpu); | |
9b51f66d IM |
366 | extern void perf_counter_init_task(struct task_struct *child); |
367 | extern void perf_counter_exit_task(struct task_struct *child); | |
0793a61d TG |
368 | extern void perf_counter_notify(struct pt_regs *regs); |
369 | extern void perf_counter_print_debug(void); | |
1b023a96 | 370 | extern void perf_counter_unthrottle(void); |
01b2838c IM |
371 | extern u64 hw_perf_save_disable(void); |
372 | extern void hw_perf_restore(u64 ctrl); | |
1d1c7ddb IM |
373 | extern int perf_counter_task_disable(void); |
374 | extern int perf_counter_task_enable(void); | |
3cbed429 PM |
375 | extern int hw_perf_group_sched_in(struct perf_counter *group_leader, |
376 | struct perf_cpu_context *cpuctx, | |
377 | struct perf_counter_context *ctx, int cpu); | |
37d81828 | 378 | extern void perf_counter_update_userpage(struct perf_counter *counter); |
5c92d124 | 379 | |
0322cd6e PZ |
380 | extern void perf_counter_output(struct perf_counter *counter, |
381 | int nmi, struct pt_regs *regs); | |
3b6f9e5c PM |
382 | /* |
383 | * Return 1 for a software counter, 0 for a hardware counter | |
384 | */ | |
385 | static inline int is_software_counter(struct perf_counter *counter) | |
386 | { | |
f4a2deb4 PZ |
387 | return !perf_event_raw(&counter->hw_event) && |
388 | perf_event_type(&counter->hw_event) != PERF_TYPE_HARDWARE; | |
3b6f9e5c PM |
389 | } |
390 | ||
b8e83514 | 391 | extern void perf_swcounter_event(u32, u64, int, struct pt_regs *); |
15dbf27c | 392 | |
0793a61d TG |
393 | #else |
394 | static inline void | |
395 | perf_counter_task_sched_in(struct task_struct *task, int cpu) { } | |
396 | static inline void | |
397 | perf_counter_task_sched_out(struct task_struct *task, int cpu) { } | |
398 | static inline void | |
399 | perf_counter_task_tick(struct task_struct *task, int cpu) { } | |
9b51f66d IM |
400 | static inline void perf_counter_init_task(struct task_struct *child) { } |
401 | static inline void perf_counter_exit_task(struct task_struct *child) { } | |
0793a61d TG |
402 | static inline void perf_counter_notify(struct pt_regs *regs) { } |
403 | static inline void perf_counter_print_debug(void) { } | |
1b023a96 | 404 | static inline void perf_counter_unthrottle(void) { } |
15dbf27c | 405 | static inline void hw_perf_restore(u64 ctrl) { } |
01b2838c | 406 | static inline u64 hw_perf_save_disable(void) { return 0; } |
1d1c7ddb IM |
407 | static inline int perf_counter_task_disable(void) { return -EINVAL; } |
408 | static inline int perf_counter_task_enable(void) { return -EINVAL; } | |
15dbf27c | 409 | |
b8e83514 | 410 | static inline void perf_swcounter_event(u32 event, u64 nr, |
15dbf27c | 411 | int nmi, struct pt_regs *regs) { } |
0793a61d TG |
412 | #endif |
413 | ||
f3dfd265 | 414 | #endif /* __KERNEL__ */ |
0793a61d | 415 | #endif /* _LINUX_PERF_COUNTER_H */ |